Illinois Crop-Acreage Estimation Experiment

Purdue University Laboratory for Applications of Remote Sensing

Purdue e-Pubs

LARS Symposia

1-1-1976

Illinois Crop-Acreage Estimation Experiment

Robert M. Ray
Harold F. Huddleston

Follow this and additional works at: http://docs.lib.purdue.edu/lars_symp

Ray, Robert M. and Huddleston, Harold F., "Illinois Crop-Acreage Estimation Experiment" (1976). LARS Symposia. Paper 167.
http://docs.lib.purdue.edu/lars_symp/167

This document has been made available through Purdue e-Pubs, a service of the Purdue University Libraries. Please contact [email protected] for
additional information.

Reprinted from

Symposium on

Machine Processing of

Remotely Sensed Data

June 29 - July 1, 1976

The Laboratory for Applications of
Remote Sensing

Purdue University
West Lafayette
Indiana

IEEE Catalog No.
76CH1103-1 MPRSD

Copyright © 1976 IEEE
The Institute of Electrical and Electronics Engineers, Inc.

Copyright © 2004 IEEE. This material is provided with permission of the IEEE. Such
permission of the IEEE does not in any way imply IEEE endorsement of any of the
products or services of the Purdue Research Foundation/University. Internal or personal
use of this material is permitted. However, permission to reprint/republish this material
for advertising or promotional purposes or for creating new collective works for resale or
redistribution must be obtained from the IEEE by writing to [email protected].

By choosing to view this document, you agree to all provisions of the copyright laws
protecting it.

ILLINOIS CROP-ACREAGE ESTIMATION EXPERIMENT*

Robert M. Ray III and Harold F. Huddleston

Center for Advanced Computation,
University of Illinois at Urbana-Champaign,

Urbana, Illinois; and

Statistical Reporting Service,
U. S. Department of Agriculture

Washington, D. C.

ABSTRACT truth information was acquired during the Illinois
1975 growing season by SRS acting in collaboration
This paper describes remote-sensing data anal- with the Illinois Cooperative Crop Reporting Ser-
ysis research conducted collaboratively during the vice. Digital data tapes for all 1975 late-summer,
last year by personnel of the Center for Advanced cloud-free LANDSAT imagery over Illinois were made
Computation (CAC) of the University of Illinois at available to the project by NASA's Office of Earth
Urbana-Champaign and the Statistical Reporting Observations Programs acting in cooperation with
Service (SRS) of the U. S. Department of Agricul- NASA's Ames Research Center.
ture. The research reported has been undertaken
by CAC and SRS to assess the practicalities of In this paper we describe the overall method-
existing high-volume earth observations data acqui- ology adopted for this investigation of the practi-
sition, processing, and communication technologies calities of LANDSAT imagery analysis for USDA
such as LANDSAT, the ILLIAC IV parallel computer, crop-acreage estimation purposes and report re-
and the ARPA Network as mechanisms for improving search findings to date. We describe the general
the accuracy of USDA annual estimates of agricul- strategy pursued in developing a comprehensive
tural crop acreages for geographic regions corre- LANDSAT imagery analysis system of the scale
sponding to U. S. states. required for monitoring agricultural crop acreages
over a geographic region of the scale of the state
INTRODUCTION of Illinois. For a region corresponding to ten
(10) western Illinois counties (a subset of the
Throughout this research, our basic apPfo~ch 102 counties of Illinois), we present preliminary
has been to seek a~ integration of ILLIAC IV ' crop-acreage estimation results derived from
and ARPA Network3 , software systems developed ILLIAC IV - ARPA Network analysis of LANDSAT data.
previously at CAC for more coss-gfficient machine Assuming the practicality of similar analyses for
interpretation of LANDSAT data' with geographic LANDSAT imagery covering the entire state, we
information systems implemented explicitly for discuss a procedure for evaluating statistically
interactive digitizing, storage, and retrieval of the information to be gained by estimating state
large quantities of crop-acreage information crop-acreage totals from LANDSAT imagery classifi-
collected routinely by SRS in the course of the cation results where SRS sample survey data are
extensive field surveys associated with its on- used as ground-truth information for classifica-
going agricultural production estimation methodol- tion training as opposed to estimating state crop-
ogy. Our primary goal has been to determine the acreage totals directly from SRS survey data alone.
extent to which SRS ground survey samples may be
employed successfully as ground-truth information GROUND DATA COLLECTION, STORAGE;
for calibrating ILLIAC IV procedures for classifi- AND RETRIEVAL
cation of LANDSAT multispectral scanner (M.SS)
imagery for regions corresponding to U. S. states. In support of this research project, all crop-
acreage information collected by SRS within the
For this exploratory application of machine state of Illinois in the course of its 1975 crop
processing of LANDSAT data, the state of Illinois and livestock surveys was retained and re-formatted
was selected as the basic study area. All ground- for use as ground-truth information for calibra-
tion of LANDSAT imagery analysis systems. These
'iii data contain complete descriptions of all agricul-
tural and non-agricultural fields, i.e., areas of
This research was supported in part by the U. S. homogeneous land cover, for all ownership tracts
Department of Agriculture through USDA Research within each of 300 area segments of the SRS
Agreement No. l2-18-04/-8-l794-X, and in part by national survey sample that fall within the state
the National Aeronautics and Space Administration of Illinois.
through NASA Grant NGR 14-005-202.

PB-14

In accordance with SRS survey procedures, task is being accomplished using graphics data
these 300 area segments had been selected earlier tablet digitizing equipment connected via the ARPA
with respect to strict statistical sampling Network to interactive DEC PDP-10 computers at
criteria and hence, while allocated heavily to Bolt, Beranek and Newman (BBN) in Boston. Data
agricultural terrains, may be considered randomly tablet digitizers at CAC are connected directly to
distributed throughout all land in the state. Each the ARPA Network through CAC's own ANTS (ARPA Net-
area segment corresponds to a geographic area of work Terminal System) computer facilities. S~S
approximately one square mile. Each segment typi- digitizing equipment has been linked to BBN comr
cally contains multiple ownership tracts with nu- puter systems via dial-up telephone line connection
merous fields ranging in size from several-acre to ARPA Network node facilities at the National
farmsteads, ponds, and forested areas to several- Bureau of Standards in Gaithersburg, Maryland and
hundred-acre agricultural fields. at Fort Belvoir, Virginia.

Following standard SRS survey practices, All agricultural field boundary digitizing is
throughout the summer of 1975 ASCS aerial photo- being accomplished using an interactive DEC PDP-IO
data tablet software system developed at CAC ex-
graphs (at a scale of 8" = 1 mile) were taken by plicitly for take-off of S'S crop-acreage data
recorded on aerial photos. This interactive data
survey enumerators to the location of each segment tablet software package was implemented as an
and used for delineation of all current field extension of the EDITOR system -- a general PDP-10
boundaries. Field boundaries for all tracts of LANDSAT imagery analysis system developed previous-
all segments were monitored continually throughout ly at CAC as an interactive ARPA Network interface
the summer in conjunction with June, July, August, to LANDSAT image interpretation procedures avail-
and September surveys conducted by SRS personnel. able on the ILLItC IV computer at NASA's Ames
Field boundary changes from month to month were Research Center.
recorded using a color-coded marking system.
These additional procedures added to the
All crop-acreage data recorded by field enu- EDITOR system for digitizing SRS crop-acreage data
merators on ASCS photos and interview forms were also include provisions for on-line geographic
re-checked independently for consistency by person- registration of all field boundaries digitized
nel of the central offices of the Illinois Coopera- with respect to USGS quadrangle map coordinates.
tive Crop Reporting Service in Springfield. All This task is done simply by mounting simultaneously
crop-acreage data contained on survey forms were both photo and quad map on the active surface of
put into machine readable format. Output from this the data tablet (36" x 48") and digitizing points
process consisted of a computer tape for which of geographic correspondence visible within both
individual records represent crop-acreage informa- the photo and quad map.
tion for all fields of all tracts in all segments
for each of the four surveys conducted throughout After digitization and geographic registration
the summer. of all segment, tract, and field boundaries delin-
eated on anyone photo, an areal-network mask is
As still another source of ground-truth infor- determined by the software system for the segment
mation, low-altitude infrared photography (at a digitized. This segment network mask is stored as
a DEC-10 disk file in terms of a list of network
scale of approximately 5" = 1 mile) was obtained nodes and links representing respectively digitized
field corners and boundaries.
commercially for a subsample of 202 area segments.
This current aerial photography provided directly Immediately following digitization and regis-
an accurate, high-resolution picture of the agri- tration of all crop-acreage data on any photo, two
cultural crops and land uses actually existing in line plotter displays are produced using a drum
late summer for the 202 segments covered. Hence, plotter at CAC to provide a hard-copy record of
it was possible to check the degree of accuracy the segment mask thus created. One of these dis-
with which 1975 field boundaries had been delin- plays is plotted at the exact scale of the photo
eated on the older ASCS photos. For those segments itself and hence, by overlaying photo and plot,
for which summer 1975 photography had been ob- the correctness of all digitized boundaries may be
tained, field boundaries (and changes) were re- conveniently checked. The other display is plotted
drawn directly on the current photography making at the scale and cartographic projection of the
reference both to data recorded by survey enumera- USGS quad map and by overlaying this plot and quad
tors on ASCS photos and to features visible direct- map the accuracy of geographic registration may be
ly in the current infrared photos themselves. This verified. (See Figures 1-2.)
task was also done in Springfield by SRS personnel.
A quantitative evaluation of the relative advan- LANDSAT IMAGERY SELECTION
tages for LANDSAT imagery classification objectives AND PREPROCESSING
of this current photography and the older ASCS
photography is currently being conducted by SRS. All LANDSAT imagery collected over Illinois
during the summer of 1975 was acquired from NASA in
To make all crop-acreage data thus compiled the form of 70 mm film transparencies and evaluated
convenient for LANDSAT imagery analysis purposes, by SRS and CAC with regard to project objectives.
all field, tract, and segment boundaries recorded Assuming ideal meteorological conditions, only
on a complete set of area segment photos (202
current infrareds and 98 ASCS photos) are presently
being digitized jointly by personnel of CAC in
Illinois and personnel of SRS in Washington. This

Ii'

I

PB-15

eleven (11) frames of LANDSAT imagery are required LANDSAT image files created, the digital image data
for complete coverage of the entire state. Given for each district is being geometrically corrected
prevailing conditions, however, a total of sixteen and geographically registered to USGS topo maps
(16) frames of imagery acquired between the dates existing for the state. This task is being per-
of 16 July and 7 September was deemed necessary to formed using an image skew transformation procedure
obtain cloud-free coverage of all of the 102 developed at CAC for efficient de-skewing and
counties within Illinois. Digital data tapes and rotation of LANDSAT digital data to map orienta-
positive film imagery (both at 1:1,000,000 and tionS in conjunction with other systems developed
1:500,000) were obtained for each of these sixteen at CAC for preci~ion geographic registration of
(16) scenes. LANDSAT imagery.

Since one of the goals of our project is to Finally, all image files are being geographi-
obtain crop-acreage estimates for the entire state cally registered to the SRS ground-truth data
of Illinois, and since counties represent smaller available (and hence simultaneously also to the
geographic units more convenient for estimation of USGS map control already associated with all
state-wide crop-acreages in terms of regional sub- ground-truth). This step is being accomplished in
totals, it was decided to preprocess and re-format the following manner.
all LANDSAT imagery acquired from NASA into a set
of image files such that each of the 102 counties First, with respect to digital image calibra-
of Illinois was contained wholly and cloud-free tion information available, all SRS area segments
within at least one image file. To accomplish are located approximately in terms of digital
this objective, the following strategy has been image file row and column coordinates. Gray-scale
adopted. displays for windows of LANDSAT data known to con-
tain all SRS area segments are produced using a
Despite the vertical overlap of approximately conventional line printer and over-printing tech-
fifteen (15) miles between successive LANDSAT niques. Then digitized SRS segment masks (de-
frames of the same orbit, in many cases individual scribed above) are again plotted this time at the
counties falling on north and south frame bound- exact scale of the line-printer LANDSAT imagery
aries are not wholly contained in anyone frame. displays. Following manual overlay and visual
Hence, in numerous instances it is necessary to correlation of line-printer and plotter displays
compile pseudo-frames of LANDSAT digital-imagery on a light table, overlay positions of maximum
by concatenating data records of a top portion of geographic correspondence between LANDSAT image
one frame to data records of a bottom portion of pixels and SRS segment masks are recorded. (See
another frame of the same orbit. Such pseudo- Figures 3-4.)
frames are to be compiled wherever they are neces-
sary to achieve continuous cloud-free imagery for LANDSAT DATA ANALYSIS SYSTEMS
a particular county.
As the SRS ground-truth data is digitized and
Due to the size of counties in Illinois, the LANDSAT imagery preprocessed for each analysis
horizontal overlap of approximately fifty miles district, LANDSAT data is being analyzed collabora-
between frames of successive orbits is sufficient tively by SRS and CAC using a common set of com-
to insure that no county fails to lie wholly with- puter facilities available via the ARPA Network.
in the swath of at least one orbit. Thus fortu- For small-scale analyses of SRS area segment data
itously, the considerably more difficult problem the EDITOR software system at BBN is used. For
of splicing LANDSAT imagery horizontally across specific large-scale LANDSAT image analysis func-
orbits does not arise. tions, the ILLIAC IV at NASA's Ames Research
Center is employed also via the ARPA Network but
Having obtained a complete set of image files addressed conveniently through the EDITOR system
(LANDSAT frames and pseudo-frames) such that each at BBN.
county is completely contained in cloud-free
fashion within at least one image file, the com- All ILLIAC IV - ARPA Network image analysis
plete set of 102 counties is to be subdivided among systems implemented to date have been designed and
non-overlapping subsets of contiguous counties, one developed in close collaboration with personnel of
group of counties per each image file. These the Laboratory for Applications of Remote Sensing
groups of counties are to be designated for project (LARS) of Purdue University. Hence all software
purposes as LANDSAT imagery analysis districts. procedures implemented specifically for machine
All subsequent data management and machine process- interpretation of LANDSAT data follow closely
ing of LANDSAT data is then to be structured in multispectral image interpretation methods re-
terms of the geographic regions corresponding to searched previously at LARS.lO,ll
these analysis districts. Inspection of the
imagery available suggests that for 1975 Illinois Specifically, ILLIAC IV procedures are now
LANDSAT imagery only fourteen (14) such analysis operational for both multivariate cluster analysis
districts -- ranging in size from as few as two or and maximum-likelihood statistical classification
t,ree counties to as many as a dozen -- are re- of LANDSAT image samples. The speed of the
quired to provide integral-county, cloud-free ILLIAC IV with respect to these two image inter-
LANDSAT coverage for the entire state. pretation procedures has proven to be generally
two orders of magnitude faster than execution
Once a comprehensive set of analysis districts times observed for the same processing tasks using
has been established and their corresponding
the IBM 360/67 computer at LARS. 5

PB-16

Such LANDSAT imagery interpretation capabili- within the ith analysis district (known
ties available via ILLIAC IV batch processing. from sampling frame)
together with the availability for classifier
training operations of the interactive image pro- a regression estimate of the average
cessing software of the EDITOR system, suggest that number of acres of the crop per area
operational crop-acreage monitoring via digital segment for the ith district
processing of orbital remote-sensing imagery may
indeed be practical. Our project has been under- the number of area segments sampled in
taken to assess more exactly the potentialities
existing in this area. In the next section. we the ith district •
present preliminary results of one LANDSAT imagery
analysis experiment using SRS data available for a average number of acres of the crop
single analysis district consisting of ten (10) reported per area segment for all n.
counties in western Illinois.
th ~
EVALUATION PROCEDURE area segments sampled in the i dis-
trict
Of central importance to our experiment is the
evaluation of the extent to which regional crop- average number of pixels classified
acreage estimates may be improved by estimation into the crop per area segment for all
with respect to LANDSAT imagery classification
methods as opposed to estimation directly from SRS ni area segments sampled in the ith
survey data alone. Statistical regression tech- district
niques may be used to obtain estimates of the total
average number of pixels classified
racreag of each crop type for each analysis dis- into the crop per segment over all

trict. 2 Following estimation of crop-acreages for possible segments for the ith district
all analysis districts separately, state-wide acre-
age estimates for each crop may easily be obtained the regression coefficient between y ..
by simply summing individual district estimates. ~J
Such estimates may be determined both with and
without use of LANDSAT classification results and and Xi' binasethdeoni ththedinsit area segments
a measure of the value of the LANDSAT data may be sampleJd rict
computed.
number of acres of the crop enumerated
Following ILLlAC IV classification of all
LANDSAT pixels contained within the counties making for the jth segment sampled in the ith
up a particular analysis district, classification district
results for each crop type will be aggregated to
obtain individual totals for all segments sampled number of pixels classified into the
within the district. Also, acreage totals for each crop for the jth segment sampled in the
crop type will be determined for the entire analy- ith district
sis district itself.
( n i
An estimator of the total acreage for a par- I:
ticular crop in a particular analysis district and
its sampling error may then be computed as follows. n i Y2 j=l
The total acreage may be estimated as
I: ----="-"----
and the variance for a large sampie of segments is: j=l ij
ni

For the individual analysis districts, the normal r.2 correlation coefficient squared be-
approximation for small samples is used, that is ~
tween Yij and xij for the ith district
V(Yi ) for large samples multiplied by (1 + n. 1 3) •
The formulas given are appropriate for a sim-
~ - ple random sample within each analysis district.
However, the SRS surveys are stratified by land use
Where Ni'l i = total acres of the crop within all .th categories which require that item totals, sums of
squares, and sums of cross products be weighted and
area segments contained within the ~ combined in order to obtain the equivalent of a
simple random sample over the entire analysis dis-
analysis district trict.

An estimate of the total state-wide acreage
for each crop may be obtained by making use of the
additive property of the estimator and its sampling
error over all districts. The estimators are

f
Y

total number of all segments contained

PB-17
ii

I.i

r

f A sample of fields was selected from the seg-
ments falling in the LANDSAT image. The acres in
V(Y) L V(Yi ) each crop or land use type was "expanded" to cor-
rect for varying probabilities of selecting seg-
i=l ments. Then a sample of fields was selected
independently for each crop so that each acre (or
A measure of the gain in relative efficiency pixel) had an equal chance of being selected for
of estimation of state-wide acreage obtained by cover types with 80 or more fields. That is, the
using machine-interpreted LANDSAT data may be com- probability of a field being selected was propor-
puted tional to its expanded acres. The selection was
made from a listing of fields ordered by segment
f Ni2 VGi ) numbers to help insure that fields would be spread
over the entire LANDSAT image. The number of
RE L fields selected for calculating mean vectors and
covariance matrices are given in tables 2 and 3.
i=l
Table 2. Number of Sample Fields by Cover Type
f Ni2 V<Yi ) (1 - r2i)n i(-~)1
i
L

i=l

where the value of RE is expected to be greater Crop or Number: Acres/: Total Nonborder
than 1.0. A value of RE less than 1.0 would indi- cover type fields: field pixels pixels
cate that information had been lost through use of
LANDSAT data. A value of 5.0 would indicate the Corn 425 23.3 9026 5604
regression estimator using LANDSAT classification Soybeans 215 22.5 4502 2712
results is equivalent to increasing the number of Perm. pasture: 163 19.7 2780 1289
area segments by five times if costs of acquiring Dense woods 144 16.8 2147
the LANDSAT data are equal to the costs of collect- Hay 11.8 1069 784
ing the area segment data. For the single analysis Wasteland 83 8.7 2087
district analyzed, the information gain or loss is: Alfalfa 274 423 477
Wheat stubble: 11.0 259
1 ~- 3 Water 40 11.2 920
Crop pasture 27 12.1 190 183
17 13.3 280
21 86
73
119

where nh is the number of area segments sampled. Table 3. Number of Training Fields by Cover Type
These values for the first analysis district are
shown in the last column of table 5.

To date limited results are available for only Crop or Number Nonborder
one analysis district of 10 counties in western cover type fields pixels
Illinois. A maximum likelihood quadratic classi-
fier using the prior probabilities for 10 land Corn 50 1648
cover categories was used for classifying each Soybeans 50 1107
pixel into one of the 10 categories for the entire Perm. pasture 25
analysis district. The prior probabilfties were Dense woods 40 297
calculated from the ground enumerated data in the Hay 16 453
10 counties. Wasteland 8 153
Alfalfa 40 492
Table 1. Prior Prob- Wheat stubble 27 183
abilities for Land Cover Categories Water 17
Crop pasture 21 86
73
119

Crop or land use Prior probabilities
(Survey land use
July 27, 1975) The pixels for all the selected fields were
combined and treated as one large field for analy-
Corn .3282 (.3097*) sis purposes; however, only the nonborder pixels
.1602 (.2297*) were used in calculating the mean vector and co-
Soybeans .1392 variance matrix. (It is planned to investigate the
Permanent pasture .0935 use of all fields and all pixels in developing mean
Dense woods .0180 vectors and covariance matrices as weil as using
Alfalfa hay .0467 equal prior probabilities in the classification.)
Other hay .0101
Wheat stubble .0118 The estimates and their errors are based on
Crop pasture .0074 the 33 segments falling in the 10 western Illinois
Water (farm ponds & lakes) .1148 counties comprising the first analysis district
Wasteland (no agri. prod.) .0701 corresponding to LANDSAT image ID#2l94-l6042 of
August 4, 1975. The estimates are shown in table 4
O~er crops & land uses and their sampling errors squared ·in table 5 for
eight agricultural land use categories. The window
(Training data not available): containing the 10 counties included 4,887,960

* Based on 1974 crop year data from Illinois

Assessor Census

PB-18

pixels and required less than 80 seconds for clas- These results for the first LANDSAT image are
sification on the ILLIAC IV. quite encouraging. Assuming LANDSAT digital tapes
and near real-time processing of the ground and
Table 4. Estimates of Agricultural Cover Types classification data, acreage estimates of spring
planted crops could have been significantly im-
Crop or Reported Regression Pixel proved for the area of this LANDSAT image by
cover type acres estimate count September 1. The authors believe that similar re-
x 1.114 sults can probably be achieved in other areas if
Corn Jull 27 (000 acres) the same conditions can be met; namely:
Soybeans 2105
Perm. pasture 1286 1390 610 (1) excellent quality, cloud-free LANDSAT
Hay 631 701 678 imagery
Alfalfa 533 434 104
Wheat stubble 179 154 14 (2) good geographic registration of ground
Water 69 0.3 segments to LANDSAT imagery
Crop pasture 39 71 10
28 39 0 (3) mean vector and covariance matrices for
45 32 each crop for each LANDSAT frame
45 (i.e., localized classifiers)

Table 5. Variances of Estimates of Agricul- (4) prior probabilities for each LANDSAT
tural Cover Types for 10-County Analysis District frame (i.e., localized priors)

Vari- Variance : Informa- (5) sufficient ground data for each crop for
tion gain classifier training
Crop or ance regression:
or loss (6) an adequate number of ground segments for
cover type : reyorted : estimate: each LANDSAT frame to compute the regres-
(1) f (2) sion and correlation coefficients for
:(100acres2):(106acres2): each crop.
(3)
(1) (2) REFERENCES

Corn 17202 2459 7.00 1. D. L. Slotnick, "The Fastest Computer," Scien-
Soybeans 5880 847 6.94 tific American, Vol. 224, No.2, February 1971,
Perm. pasture 4489 1096 4.09 pp. 76-87.
Hay 630 376 1.67
155 135 1.14 2. W. J. Bouknight, S. A. Denenberg, D. E.
Alfalfa 66 McIntyre, J. M. Randall, A. H. Sameh, and D. L.
Wheat stubble 30 70 .94 Slotnick, "The ILLIAC IV System," Proceedings
Water 88 11 2.71 of the IEEE, Vol. 60, No.4, April 1972,
Crop pasture 94 pp. 369-388.
.94
3. L. G. Roberts and B. D. Wessler, "Computer Net-
SUMMARY work Development to Achieve Resource Sharing,"
1970 Spring Joint Computer Conference, AFIPS
The four dimensional distributions by cover Conference Proceedings, Spartan, 1970, pp. 543-
types exhibited considerable overlap except for 549.
water. In general, the distributions were unimodal
or where several modes were present, one mode was 4. L. G. Roberts, "Network Rationale: A 5-Year
much higher than the other. Soybeans had two dis- Reevaluation," COMPCON 73 Proceedings, Seventh
tinct modes (one major and one minor), but no Annual IEEE Computer Society International Con-
factor could be isolated as a basis for grouping ference, March 1973, pp. 3-5.
fields into two categories. Based on the principal
component analysis, bands 6 and 7 (IR) explain 5. Robert M. Ray III, John D. Thomas, Walter E.
practically all the variation for most cover types, Donovan, and Philip H. Swain, "Implementation
but for several cover types bands 4 and 5 (visible) of ILLlAC IV Algorithms for Multispectral Image
were equally important. The quality of the data Interpretation," CAC Document No. 112, (June
for band 7 appeared to be superior to the other 1974), Center for Advanced Computation, Univer-
bands which had discontinuities or gaps in the data sity of Illinois at Urbana-Champaign, Urbana,
values. Illinois 61801.

The LANDSAT data for major crops or cover 6. Robert M. Ray III, Martin Ozga, Walter E.
types (i.e., large acreages) showed significant im- Donovan, John D. Thomas, and Marvin L. Graham
provements can be expected in the estimates but "EDITOR: An Interactive Interface to ILLIAC '
minor cover types showed little improvement'or even IV - ARPA Network Multispectral Image Process-
loss of information unless their distributions were ing Systems," CAC Document No. 114, (June 1975),
reasonably separated (i.e., in the measurement Center for Advanced Computation, University of
space) from the distributions of the major cover
types.

PB-19

Illinois at Urbana-Champaign, Urbana, Illinois ? ~
61801.
+ +""
7. Walt Donovan and Martin Ozga, "Retrieval of
LANDSAT Image Samples by Digitized Polygonal +
Windows and Associated Ground Data Informa-
tion," CAC Technical Memorandum No. 57, (August !:.,172
1975), Center for Advanced Computation, Univer- Sr:C"1!:NT SCALr: IS 1 1203[;
sity of Illinois at Urbana-Champaign, Urbana,
Illinois 61801. ~igure 2. Example Area Segment Mask Plotted at
Scale of USGS 15' Quad and Showing 5' Quad Ticks.
8. Walt Donovan, "Oblique Transformation of ERTS
Images to Approximate North-South Orienta-
tion," CAC Technical Memorandum No. 38,
(November 1974), Center for Advanced Computa-
tion, University of Illinois at Urbana-
Champaign, Urbana, Illinois 61801.

9. Walter E. Donovan, Martin Ozga, and Robert M.
Ray III, "Compilation and Geographic Registra-
tion of ERTS Mu1titempora1 Imagery," CAC
Technical Memorandum No. 52, (May 1975), Center
for Advanced Computation, University of
Illinois at Urbana-Champaign, Urbana, Illinois
61801.

10. "Remote Multispectral Sensing in Agriculture,"
Research Bulletin 873, (December 1970), Labora-
tory for Agricultural Remote Sensing, Purdue
University, West Lafayette, Indiana 47907.

11. P. H. Swain, "Pattern Recognition: A Basis for
Remote Sensing Data Analysis," LARS Information
Note 11572, (1972), Laboratory for Applications
of Remote Sensing, Purdue University, West
Lafayette, Indiana 47907.

12. W. G. Cochran, "Sampling Techniques," (2nd Ed.)
(1963), Wiley and Sons, pp. 193-200.

~

,~-----.--------

I,

I

~~~__~___~____~~___J

I S172 .

.sr:6'1":NT SC..6..U: 1S 1; 12e.Jt~

Figure 1. Example USDA/SRS Area Segment Mask Plotted
at Scale of Photo Digitized. (Shown reduced here.)

PB-20

33333333333333333333333333333533333333 ~------------------------------------~
22222222222222222222222233333333333333
77778888888888999999999900000000001111 ~
678901234S678901234S678901234S67890123 M
M
627' • •• ~w~.G··;"S5· "~;;"
630

628 S- a;:iGSiGli ;81"- -,,;;"
629 •• ';;;;;IIJ 'BI&S' ;;;";
630 •.• ······Sl ··GG•.•oSa·~·"';···"'···';;~"~SG';
631 II'I'''G:GGWSI'
""'':1,'111111~~~ :111::I·III·:=:I~~:~~~'~:·t'
···,,'G........ I634 • ·
• •
63S
636
• •• ,.G; •••a• o.. I II· ·."·S'"·;;'GWwWw~W~O.;a. ', :·:··
I •• ~····";W •••
• S• •

6~3387 "8~8O"0O.0O'8"1"";"·"·'·0·"·.~G6•"1.•"0•e0".8O. ..".. . ' •. •.•.''G;'"'G'G'GOiOlwGW8a..'GGl'GG' 0;8' i
639
. . ~
640 80S0.G·· ···; •••• SS ••••••• ";WWO •• GGGG
641 "GGOG" ···,••••••••• a••• ";a.e.SGGWG i
642 Gil . . . . '" ;GII.II ••••• aa .. ;'·'" 'G ••.• GGGG
643 WGO.......... 100Wil~G' ·Gill.;G;G !~
644 ;;O •• laSlsaa ••• sOSGWGGi· ';ilO'WGGG
645 GG •••• OG·,,'OG;"·'ilGG .. '· ·GiI •••• OO
646 G•••• OS;,;;;;"'··' •• 5G; ·W"'.S"'O'
647 . . . . . O;·,;;;;,·,;;;GGS.•• ;o . . . . . . . . . e ••

G.648 •••• 0' ··,;GWWGwWG88Ge ••••••• W••••••

649 ••••• ' ··'GO .. OOIIG .. SOGG •• , ••••• ;G.;···,
6S0 ••••• ; ••• ,; ... GS.WGGIIG ...... 'II.G •• , . . :

"11' ..665512 ••••••••;·G····;·S,G'SGOw.a.w..wwwBBOGIGG;Oa.S.·,1 • •• I.;b
653 GG.WGWOO.;JO.IWG........ • •• e•••••w.
654 ·666Ge9 •• 0••'. iG•••, ••e •••••ee••••••••
655 WGWI.SGG .. eilS •••••••••I•,__ iIwS W....SaJW. GSS.SI8Ww' , . I ."•' 10 1_••
656 .GGG;G;;G;Gw i

: SEGf1El\,"f 5236. PART 1 ()~ 1. 33 ~rt:LDS 281 EDGES 249 \oEQrIcr:s

TOTAl. ARE~ 657.2 ~C'<;:S. SC.~i..C ;s 1. 12122

PLO; F"Li...E (Ii_:LrM)rs)~238.Sr:G;7.22.....,I\R·76 06.54.21

USER is ,..010t, 01' •• gfflent.. u.~"g ,ocel eel ~c,..t.~on

E~UIPROBAaLE SCALING USED. CAl. f":LE (ILLlNOIS><;236.C.~; 1.2S-,1.R-76 13."8.31

Figure 3. Gray-scale Line-printer Display of Figure 4. Plotter Display of Same Area Segment
Digital LANDSAT Data (IR Band 4) Corresponding to Scaled Horizontally and Vertically to OVerlay Line-
a Specific.USDA/SRS Area Segment. printer Display for Precision Local Registration of
Ground Truth Data and LANDSAT Imagery.

PB-21